
210 ◾ Transportation Statistics and Microsimulation
deviance/DF should be close to 1 if the model ts the data well. A value of
deviance/DF much greater than 1 suggests that the model is misspecied
or there is an overdispersion problem in the data. A value of deviance/
DF much less than 1 also indicates misspecication of the model or an
underdispersion problem (the phenomenon that the variance is less than
the mean, which also violates an underlying assumption for the Poisson
distribution) in the data (Sheather, 2009). Caution needs to be exercised
with using the deviance as a goodness-of-t measure when the data set
being analyzed is ...